AI bots are using credit cards. Banks are racing to keep up.

Summary

Stripe is positioning itself as a key cog for agentic commerce, while JPMorgan Chase is a skeptic.

Sudden advancements by artificial intelligence bots are opening new fronts in the battles over who will control the future of consumer payments.

Meta's Muse, Instinct and other personal AI assistants have been empowered in recent weeks to make purchases on behalf of cardholders, marking new steps toward a potential future where consumers no longer need to initiate a purchase or press "buy" at checkout. That is generating buzz and excitement about AI bots' ability to create a new world order in payments.

But inside banks and other payments companies and merchants, worries abound about the implications and whether bots will lead to elevated cases of fraud and returns, and open new legal risks. Early camps are already being formed.

Payments giant Stripe has embraced agentic purchasing and sought to stake out a central position in the AI future, according to people familiar with the matter. Cards from banks such as Bank of America, Capital One and Citigroup are being used by agents when consumers enable them.

JPMorgan Chase, the largest U.S. credit-card issuer by purchase volume, is among the more skeptical banks, people familiar with the matter said. The bank has been blocking agents when consumers give the bots their card information and it has been slow to allow other solutions rival banks are trying, the people said.

It is early days, and the number of consumer card transactions made by AI bots is small, but it is likely to grow materially, according to payments executives. The underlying tension for banks and other payments companies is the risk of being left behind if they choose the wrong technology or are too slow to act.

"It's a land grab right now," said Farooq Malik, CEO of Rain, an agentic payments and stablecoin infrastructure company.

The systems that power payments today are built on byzantine rules, fraud protections, federal laws and network and bank fees that aren't intended for robots. This ecosystem has long been in a state of constant tension between merchants, networks that set fees and rules, and card issuers that extend credit to consumers.

Move one piece or change one fee even by a cent and the impact can quickly ripple into billions of dollars won and lost.

The advancement of AI into finance is creating several concerns.

Large banks, including JPMorgan, Citigroup, Bank of America and Capital One, recently moved to restrict AI agents from accessing consumer deposit accounts over the safety of consumers' personal information. Economists and analysts have raised fears that AI agents will easily find higher-yielding deposit accounts elsewhere and transfer their deposit holders' money there.

The Treasury Department, meanwhile, is asking banks questions related to frameworks around AI and payments, according to people familiar with the matter.

In payments, banks are concerned about consumers placing their credit cards directly into the hands of AI bots. JPMorgan, for instance, has moved to restrict agents from shopping with those consumers' cards.

But executives at other large banks say they view it as a difficult-to-impossible challenge to spot AI bots and shut them off.

Michael Spelfogel, president of Cardless, a technology company in the card industry, said his company is working with network partners to enable agentic spending.

"People already want AI agents to shop for them, and the technology to do it safely is built," Spelfogel said. "Banks can slow this down for a year or two, but they can't stop people from using agents to spend their own money."

Some payment executives say they are already seeing higher returns from agentic shopping compared with when consumers make their own purchases, as well as elevated fraud cases, such as from agents purchasing items from fake online stores that never deliver the products. And there are concerns about AI supercharging points gaming, which could make premium cards more expensive for banks.

Executives are also wary of liability. Credit cards come with fraud protection for users, but banks don't view an authorized charge the same way. Some executives say their banks plan to tell consumers they are treating their authorization of an agent to shop with their card as equivalent to cardholders giving their cards to relatives or friends.

Stripe is seeking to play a key role by leveraging its Link wallet, which has more than 300 million users, to enable agentic shoppers to execute purchases.

"Agentic commerce is nascent but promising," a Stripe spokesman said, adding it is working with partners to build the technology. Some banks see Link as a preferable alternative to cardholders giving agents their raw card numbers.

Link is using agentic tokens, which are meant to make transactions more secure by hiding the card's number, similar to the tokens that hide consumers' card credentials on Apple Pay and other wallets. The agentic tokens, which are provided by Visa and Mastercard, are also packaged with other data, such as information on what consumers want purchased on their behalf, the limits they place on purchase amounts, and whether the AI bot is trusted.

That hasn't been enough for JPMorgan, which has been at odds with the agentic token technology used by Stripe and has been slow to implement it, according to people familiar with the matter.

Stripe has a separate mechanism for AI bots, which can be used for Chase cards and other issuers. In those cases, Stripe's wallet has been using one-time virtual cards that are equal to the purchase amount. Just after the transaction goes through, Stripe charges the purchase amount to the consumer's actual card, which effectively pays off the virtual card.

This arrangement has opened its own set of issues, especially for merchants. Merchants are complaining to Visa and Mastercard that they are incurring higher fees on average from the virtual card than they would be charged by the consumer's actual card, according to people familiar with the matter. That is because the virtual card is classified as a commercial card rather than a consumer card.

Merchants already have other concerns about agentic commerce, including less visibility into how their customers are shopping, which makes it harder to collect data and make future purchase recommendations to them. The temporary cards have also created challenges for processing returns, some merchant executives said.

United Airlines said Wednesday that it will prevent personal AI assistants from accessing its systems to book tickets. Amazon in September blocked Meta's Muse from making purchases. Walmart, however, said it would allow consumers to authorize AI agents and distinguish between the purchases they and the agents initiated.

The Office of the Comptroller of the Currency this month gave its own bank examiners a new framework for evaluating banks' use of AI, including payments, customer service and fraud detection. The document lists "autonomous payment agent[s]" as being at the highest risk tier.